PICI/Knowledge Graph

30 Sep 2025 - 18 Nov 2025
Open in Logseq
    • Project to do somethig with the scholarly literature
    • Enjun's collection Box | Login
      • this was part of an intern project which nobody looped me in on, I feel slightly butthurt
    • Oct 28th, 2025 talked with EnJun, he seems OK with this, there is this weird thing where it has to be his idea, which, OK, as long as I get paid and get to do something not entirely stupid, who cares. It's just weird.
      • Process the PICI grant documents (about 1000).
      • Answer questions from comms, eg, who do we have working on BRCA1 for non breast cancer
      • 3 different knowledge sources
        • grants
        • sci lit
        • general LLM
    • Nov 18th, 2025
      • EnJun wants this prioritized; it might help with the organization chaos. Ira is trying to standardize work that the network scientists do:
        • Will be great. Um, I think, because right now, what? What's also happening is, uh, Ira is now president research. He's going around all the sites to start, like talking, remember what they are. Current research and then sort of standardizing across the sites a bit more so that there's less. Supposedly, there's going to be less waste and less redundancy, but also a bit more collaboration is when he tries to build together. So, I think that's why. That's why I'm saying that maybe this system might be getting more and more...
        • I don't think here's an idea about standardization is, but he was very recently. And you know he's been going to the sites listening to them about, like, what their research is, what they're doing, and then he's identified certain sets of redundancies and and Models, and at on occasions where, like, we've been like, why are you using this? Um, like, you know, set of genes or this set of, like, uh? The definitions or cells. When internally, your own site also uses this other cell definitions for cells, or this is like, can you do that, and so I think Ira is very focused on trying to like clean that up even further. Okay? Make it a bit more unified across Pisces, but honestly, he hasn't really voiced his vision yet because I think he's just being. Careful, uh, to me, all the sites and hear what they're doing and figure out what what that is, uh, that Pisces needs to do as well.
      • Needs to be conversational, or at least that is EnJun vision.
    • Requirements

      • Ingest PICI grant proposals,
      • make them searchable or queryable, or otherwise make the information they contain more useful.
    • Use cases / user cases

      • Ira (president of research): would like to know what labs are dealing with what research items.
      • Kate (COO): Same but at a higher level, less scientific detail
      • EnJun: (Ask him)
    • UI design

      • Conversational, this is like an LLM with customized context.
        • In fact should do a standard RAG just for a first pass.
      • Idea: custom chart of labs vs research areas, where the latter can be defined by the user (eg, different genes, cancers, whatever).
      • Hah: how about a dot graph where size ≡ funding? They would eat that up, or they should.
    • Questions

      • Should this be just PICI proposals, or try to integrate scientific literature as well?
        • I guess start with the proposals, that's clearly the right thing.
      • Entity identification / extraction.
        • How to do this from text
        • Do we need to build a knowledge graph? Certainly doing so would be my instinct.
    • Plan

      • explore use of OpenAI default RAG system
      • ingest and index PICI proposal corpus into current KG rag system.
        • generalize current machinery
        • deal with docx and other file issues
      • conversational interface. Seems like there must be componentry for this. Or build one in way, I have almost all the pieces I need.
      • entity extraction, ontology, knowledge graph
      • experiment with integrating into UI (eg highlighting, filtering, graphs)